detection and classification of buried radioactive metal objects using wideband emi data

17
Electrical & Computer Engineering Anish Turlapaty, Jenny Du, and Nicolas Younan Department of Electrical and Computer Engineering Mississippi State University IGARSS 2011 Vancouver, Canada Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Upload: kiara

Post on 24-Feb-2016

45 views

Category:

Documents


0 download

DESCRIPTION

Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data. Anish Turlapaty , Jenny Du, and Nicolas Younan Department of Electrical and Computer Engineering Mississippi State University IGARSS 2011 Vancouver, Canada. Outline. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Anish Turlapaty, Jenny Du, and Nicolas YounanDepartment of Electrical and Computer Engineering

Mississippi State University

IGARSS 2011 Vancouver, Canada

Detection and Classification of Buried Radioactive Metal Objects

Using Wideband EMI Data

Page 2: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Outline• Overview and Background• Detection Methods• Methodology

– Feature Extraction– Multistage Learning– Validation

• Performance• Summary

Page 3: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Overview

EMI Spectroscopy

• Object exposed to EM Field • Generates secondary magnetic field • Measure its spectrum • Identification of the target’s finger print • Classify the EMI Response• Validate the methodology by measuring correlation

against a library of signatures from lab experiments

Goal: Identify the target (Depleted Uranium DU)

Page 4: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Background• Depleted Uranium (DU), in general, is considered both a

toxic and radioactive hazard - Effectively detecting DU is of great importance

• Although DU and other radioactive materials have different characteristics, the detection of DU from other metals is challenging due to spectrum similarities

• In practice, the situation can be much more complicated due to the presence of background clutters, especially when the DU is buried

• Even more challenging to accomplish detection in an automated fashion

Page 5: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Detection Methods• Common techniques applicable to landmine

detection– Use a library of signatures– Bayesian approach– Bivariate Gaussian model

• However the problem of variable orientation of the target object is not solved

Utilizing EMI data, a pattern recognition approach based on a decision tree for DU detection is developed

Page 6: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Validation

Feature Extraction

Multi-stage Learning

Signature Extraction

Field dataFeature Vectors

Classification Map

Best Fit

Target Signature

Methodology

Page 7: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Field Data DescriptionField Data (raw test

data)Collected from a

rectangular grid of size 60m x 16m

EMIR is collected for 7 frequencies (widely separated)

330 990 3030 6030 13050 21300 43080

71 to

Classes 1- DU at surface, 2- DU at 30cm, 3- DU at 60cm, 4- clutter

Page 8: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

MotivationQuadratic component of EMI response of 1 inch DU rods has a peak at around 3.5kHz The relation between the peak value and the values from its neighboring bands can be very useful for characterizing DU metal of the same radii.

Feature Selection Relevant features: Four spectral values of quadratic component centered around the peak value in the region of interest are

ω1

ω2

ω3 ω4

ω5

ω6

ω75432 ,,,

Page 9: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Central Features >

thr1

Central Features > other Features Background

Other Metals

Feature Vectors

YESNO

YESNO

Feature Subset

One-Class SVM Training

Map Generation

PDF Estimation

Clustering

Class MapClass

Separation

PDF Visualization

Multi-stage ClassificationCentral features are the spectral values at 43 ,

Threshold thr1 is determined from the histogram of the corresponding feature

Page 10: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

PDF Estimation

• The PDF has two distinct Gaussian peaks • The Gaussian peaks correspond to the centers of two Gaussian

distributions, thus two clusters

Page 11: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Clustering

• Individual feature vectors are classified to one of these two clusters based on their distance to the two centers

• A threshold value is used to reject vectors that do not belong to either cluster

Page 12: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Classification Map

Page 13: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Validation

Linear Model

Laboratory Measurements with

GEM-3 sensor

reftest QaQ

Reference SignaturesBest fit

Target Response

EMI Response is measured for seven metal cylindrical rods of 4inch length and 1inch diameter at 29 frequencies from 90 Hz to 90KHz.

Target response is basically quadratic response of objects at selected locations from classification map

Page 14: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Validation Contd.• The Quadratic component of the EMI

Response of the buried target should have high correlation with the EMI response of the same object measured in the laboratory.

• Depth vs. Magnitude of EMI Response for compact objects

• The magnitude of H field inversely depends on Nth power of the distance

• Thus objects of class 1 have higher magnitude as they are closer to surface

• Class 2 objects are much deeper thus relatively weaker signal strength

fQuadafQuad labfield

}){,,,(1secondary jYXzyxafH

Corresponds to Quadrature response of DU objects at the surface (Class 1)

Page 15: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Mean EMI ResponseCorresponds to DU objects at 30cm depthSupported by the reduced magnitude of the EMI response

Non-DU metal object with different EMI signature (clutter)

Page 16: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Performance

• Multistage approach with SVM for DU discrimination and depth separation

• Confusion Matrix

Predicted ClassesActual Classes

1(DU) 2(Clutter) 3(soil)

1(DU) 261 0 292(Clutter) 12 89 03(Soil) 0 0 8158

DU ObjectsClutter

Soil Background

Average accuracy 95 %

Page 17: Detection and Classification of Buried Radioactive Metal Objects Using Wideband EMI Data

Electrical & Computer Engineering

Summary

• Unsupervised classification of different objects at different depths using a multistage learning method with OCSVMs is shown to be quite successful

• This method is validated by comparing the target signatures from each class with laboratory measurements

• Extension of this work is testing the detection and/or discrimination algorithm in the presence of substantial clutter and variable size DU objects